Settings used for saving the entire TiddlyWiki as a single file via a saver module
These settings are only used when saving to GitHub
| Username | |
| Password, OAUTH token, or personal access token (see GitHub help page for details) | |
Target repository (e.g. Jermolene/TiddlyWiki5) | |
| Target branch for saving | |
Path to target file (e.g. /wiki/) | |
Filename of target file (e.g. index.html) | |
| Server API URL |
Previous:
- RADE
- On Fair Selection in the Presence of Implicit Variance
- NLP - Human In The Loop
- Comparing Human and Machine Errors in Conversational Speech Transcription
Now:
- From ImageNet to Image Classification: Contextualizing Progress on Benchmarks
- Understanding Generalization through Visualizations
- Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks
- A Closer Look at Memorization in Deep Networks
- Confident Learning: Estimating Uncertainty in Dataset Labels
- What do Bias Measures Measure
- Adversarial Examples Are Not Bugs, They Are Features
- BREEDS: Benchmarks for Subpopulation Shift
- WILDS: A Benchmark of in-the-Wild Distribution Shifts
- Robustness May Be at Odds with Accuracy <โ this one highlights the possibility that making accuracy the center of ML algorithms evaluation end up struggling with the real-world kind of evaluation we need
We still have a limited understanding of how to train the CNN models with the label noise inherent in existing face recognition datasets. To address this issue, this paper develops a novel co-mining strategy to effectively train on the datasets with noisy labels. Specifically, we simultaneously use the loss values as the cue to detect noisy labels, exchange the high confidence clean faces to alleviate the errors accumulated issue caused by the sample-selection bias, and re-weight the predicted clean faces to make them dominate the discriminative model training in a mini-batch fashion.
Methodology
Noisy Labels Detection
Intuitively, when labels are correct, small-loss instances are more likely to be the ones which are correctly labeled. Therefore, if we train our classifier only using small-loss instances in each minibatch data, it should be resistant to noisy labels. Assume that we have estimated the noise rate of a face recognition dataset. is the mini-batch size. Our method maintains two networks simultaneously. In each mini-batch of data, each peer network views its small-loss instances as the useful knowledge and drops about numbers of big-loss instances as the distractors, leaving the rest of samples into two parts, intersected faces and non-intersected faces of these two peer networks. For the intersected faces, since two peer networks predict them as clean faces, we have reason to believe that they are clean enough for deep face recognition. For the non intersected ones, they have high confidence to be clean faces.
Welcome to TiddlyWiki and the TiddlyWiki community
Before you start storing important information in TiddlyWiki it is vital to make sure that you can reliably save changes. See https://tiddlywiki.com/#GettingStarted for details
Set up this TiddlyWiki
| Title of this TiddlyWiki | |
| Subtitle | |
| Default tiddlers | Choose which tiddlers are displayed at startup Use [[double square brackets]] for titles with spaces. Or you can choose to |
See the control panel for more options.